Professional Knowledge Mining

Field Guide for the 18-Month Experiment

Professional Knowledge Mining is the practice of finding valuable human expertise, earning access to it, capturing judgment in the context of real work, protecting what you collect, refining it into a usable knowledge asset, and applying it to something that improves performance or creates economic value.

The opportunity is not mainly about collecting more information. Manuals, videos, specifications, and public AI systems already contain enormous amounts of information. The more interesting target is the judgment that experienced people use when the manual is incomplete, the situation is ambiguous, or the obvious answer is wrong.

The working sequence is:

Prospect → Partner → Capture → Protect → Refine → Apply → Test → Repeat

The first experiment teaches you a domain. The deeper goal is to learn the machinery well enough that the second experiment goes faster.

1. The Experiment in One Page

Give yourself two or three summers, or roughly 18 months of sustained effort.

Prospect: Find a domain where expert judgment matters, mistakes are expensive, and important knowledge is poorly captured.
Partner: Find one or two unusually capable practitioners and create a relationship worth maintaining.
Agree on rights and incentives: Decide what can be captured, who owns what, how the expert benefits, and what is off limits.
Embed yourself in the work: Observe ordinary jobs, difficult jobs, mistakes, exceptions, and follow-up outcomes.
Capture consequential moments: Record what happened, what the expert noticed, what they believed, what they did, and what followed.
Protect the material: Separate public, sensitive, proprietary, and crown-jewel knowledge before deciding where it can be stored or processed.
Refine the captures into cases: Create structured episodes that can be retrieved, compared, reviewed, and traced back to source evidence.
Apply the knowledge: Use it for training, diagnosis, decision support, inspection, services, tools, or another concrete job.
Test transfer and willingness to pay: See whether the knowledge improves decisions and whether anybody values that improvement enough to pay.
Run a second small experiment: Take your method into another domain and see which parts of your knowledge-mining machinery transfer.

Do not expect the sequence to run neatly from left to right. You will loop constantly. A failed case changes how you capture. A disagreement changes how you structure records. A customer request changes what you prospect for next.

That iteration is not noise around the experiment. It is the experiment.

2. Prospect for Judgment, Not Interesting People

A fascinating expert does not automatically create a strong knowledge-mining opportunity.

The best prospects tend to share several characteristics:

Important expertise is difficult to articulate.
Mistakes, downtime, rework, or slow learning are expensive.
Similar situations recur often enough for captured knowledge to be reused.
The work can be observed closely enough to connect decisions to evidence.
Outcomes can eventually be checked.
Experienced practitioners are scarce, retiring, or difficult to replace.
Public information and general-purpose AI do not already solve the problem well.

Promising areas include industrial maintenance, specialized equipment repair, precision manufacturing, utilities, legacy systems, scientific instruments, municipal systems, construction specialties, restoration, agriculture, and rare crafts.

Do not assume the biggest industry creates the best first experiment. Pipe-organ tuning could be a better starting point than industrial maintenance if you can gain extraordinary access to the right practitioner and repeatedly observe meaningful decisions.

A useful test is:

Does this expert know something expensive that is missing from the manual, and will other people encounter similar situations again?

If yes, investigate further.

3. Partner With the Expert

The human relationship is not a soft issue sitting beside the technical work. It is part of the method.

A knowledge miner who cannot earn long-term trust will collect shallow material.

You may be asking for access to decades of experience, including mistakes, shortcuts, unusual cases, customer situations, private judgments, and methods the expert may never have consciously explained. That is a substantial request.

Make the Exchange Worthwhile

Possible incentives include:

hourly or project compensation;
useful help inside the business;
organizing old photographs, notes, or records;
creating training for younger workers;
attribution and professional credit;
royalties or shared commercial rights;
ownership in a resulting product or company;
an ongoing advisory or teaching role.

Money is only one motivation. Some experts care about preserving the craft. Some want to improve the next generation. Some enjoy the intellectual challenge. Some want their name attached to a standard. Some want to solve a painful business problem.

Ask what matters to them.

Treat the Expert as a Co-Author

Bring the material back.

Show the expert what you think you learned and invite correction. Let them challenge your classification, rewrite your explanation, and identify missing context.

A particularly good sign is when the expert begins prospecting for you:

Here's one we should capture.

At that point the expert is no longer merely a source. They are participating in the knowledge-mining system.

Mine the Knowledge, Not the Person

The phrase matters.

Do not approach a practitioner as a deposit to exhaust and abandon. The relationship is part of keeping the knowledge accurate, current, and alive.

A strong arrangement should still make sense if the project becomes unexpectedly valuable.

4. Capture Judgment in the Work

The strongest knowledge miner is closer to an apprentice than an interviewer.

Much of the important material appears only while the work is happening: a pause, a sound, a change of tool, a rejected diagnosis, a tiny adjustment, a decision to leave something alone.

Start With Proximity

Take on useful tasks within your competence. Prepare materials. Arrange tools. Document jobs. Help with setup. Learn terminology. Understand enough of the routine that departures from it begin to stand out.

Your ignorance is useful at first because you ask questions insiders stopped asking years ago. Eventually, though, ignorance becomes a handicap. You need enough domain fluency to recognize when something consequential just happened.

Ask Short Questions at Meaningful Moments

Useful prompts include:

What changed?
What did you notice?
What are you ruling out?
How sure are you?
What would change your mind?
What would make you stop?
What do you expect to happen next?
What did you notice that I didn't?

Do not force continuous narration. If the work demands concentration, accept “ask me later.”

Predict Before the Outcome

Once you know enough to form a view, state it before the expert acts.

I think you're going to adjust this because of the sound.

Let the expert correct you.

This exposes missing cues and reduces the temptation to produce polished explanations after everyone already knows the answer.

Let the Expert Correct Your Work

On safe, appropriate tasks, try the work yourself.

An interruption such as:

Stop. You're correcting the wrong problem.

may expose more expert judgment than another hour of interviewing.

Ask what revealed the mistake and how early it became visible.

Replay Important Moments

Review short clips together.

You paused here, listened again, and changed tools. What happened?

Keep separate:

what the recording visibly establishes;
what the expert remembers;
what measurements show;
what remains interpretation.

Reconstruct Difficult Historical Cases

Not every important situation will occur during your residency.

Ask about near failures, unusual repairs, major mistakes, unexpected successes, or cases another expert questioned. Reconstruct what was known at each point in time rather than letting hindsight flatten the story.

Compare Nearly Identical Examples

Ask the master to compare two examples that appear equivalent to you.

Show me where the difference lives.

Then ask where the boundary lies:

How much would this have to change before you made a different decision?

Let Experts Disagree

Two excellent practitioners may reach different conclusions.

Do not automatically turn disagreement into one sanitized rule. Investigate whether the difference comes from observation, goals, experience, risk tolerance, quality standards, or different schools of practice.

The disagreement itself may be valuable knowledge.

Follow the Outcome

Return later.

Did the repair hold? Was the diagnosis correct? Did the customer remain satisfied? What surprised the expert? What would they now do differently?

A convincing explanation immediately after the work is less valuable than a judgment tied to a durable outcome.

5. Instrument Only What Matters

A wearable camera captures only part of expertise.

Some judgment depends on sound, force, vibration, temperature, resistance, timing, pressure, position, or other signals that ordinary video cannot preserve well.

Ask:

When you say you can feel or hear the difference, what could we measure alongside that experience?

Depending on the domain, useful tools may include:

thermal cameras;
vibration sensors;
sound meters;
force measurement;
moisture meters;
borescopes;
digital microscopes;
calipers;
environmental sensors;
machine telemetry;
instruments already used by the practitioner.

The goal is not to instrument everything.

Measure signals that appear to influence consequential decisions.

This becomes increasingly important if the knowledge will eventually guide machines, because descriptions alone do not reproduce physical action. Demonstrations, sensing, movement, contact, and outcome verification matter too.

6. Protect the Knowledge Before You Process It

Treat protection as part of Professional Knowledge Mining, not paperwork you handle later.

Create a Rights Map

Before serious capture begins, establish who may own or control:

raw video and audio;
photographs and measurements;
customer records;
drawings and manuals;
expert-created material;
derived case records;
software or models built from the material;
commercial usage rights;
rights to continue using material if the relationship ends.

The expert may not personally own everything they reveal. Employers, customers, manufacturers, institutions, licensors, or other parties may have rights or confidentiality expectations.

Get appropriate agreements and professional advice before commercializing anything important.

Create Sensitivity Levels

Do not treat all material the same.

A simple starting scheme might be:

Public: manuals, published materials, or information safe for ordinary use.
Internal: useful working material that should stay inside the project.
Confidential: customer-identifying, proprietary, or contractually restricted information.
Crown jewel: rare judgment, proprietary cases, or derived assets whose value depends on tight control.

The exact labels matter less than the habit of deciding before processing.

Be Careful With AI

AI can help enormously with transcription, indexing, tagging, comparison, contradiction detection, retrieval, case generation, and later products.

But uploading information to an AI service is itself a disclosure decision.

Before providing valuable material to any system, understand:

what gets retained;
whether humans can review it;
whether it can be used to train or improve models;
what contractual protections apply;
whether it can be deleted;
where it is stored;
whether the expert or customer agreements permit the use.

Some material may be fine for ordinary AI tools. Some should be anonymized first. Some may belong only in controlled systems. Crown-jewel knowledge may justify much stricter infrastructure.

Protect the knowledge layer before building the intelligence layer.

Preserve Provenance

Every important derived case should point back to its source evidence.

Know which expert, job, recording, image, document, measurement, or later outcome supports the conclusion.

Without provenance, you will eventually lose the ability to:

verify a case;
correct it;
understand its limits;
resolve disagreement;
prove where it came from;
determine whether you have the right to use it.

Protection is not only secrecy. It is control and traceability.

7. Refine Captures Into Cases

A hundred hours of video is not automatically knowledge.

Give every meaningful episode an identity.

A useful case might include:

Context: What situation was the expert facing?
Evidence: What could be seen, heard, measured, or known at the time?
Cues: What did the expert notice?
Judgment: What interpretation or decision followed?
Alternatives: What else was considered?
Action: What did the expert do?
Boundaries: When would this reasoning fail or cease to apply?
Outcome: What happened afterward?
Source: Which recordings, measurements, images, or documents support the record?
Review: Did the expert verify it? Did another expert disagree?

You can begin with simple infrastructure: secure folders, consistent filenames, transcripts, notes, a spreadsheet or database, and links back to source evidence.

Later you may add semantic search, vector retrieval, knowledge graphs, multimodal models, or specialized applications.

Do not begin there merely because those tools sound sophisticated.

A well-structured collection of 200 important cases may be far more valuable than 20 terabytes of badly organized recordings.

8. Apply the Knowledge to a Real Job

A good knowledge asset changes somebody's behavior.

Ask:

Who makes an expensive decision today that becomes easier with this judgment?

Possible applications include:

Expert Diagnosis

Customers submit images, recordings, measurements, and operating history. The business helps determine likely causes and the next diagnostic step.

Practitioner Assistance

A field assistant helps workers recognize conditions, recall similar cases, consider alternatives, and know when to escalate.

Training

Learners work through real cases, make decisions, explain their reasoning, and compare it with expert judgments and real outcomes.

Specialist Service Company

Use the captured knowledge internally to operate an actual service business better than competitors.

Quality Assurance

Use accumulated cases to identify defects, evaluate work, or verify that an outcome meets a meaningful standard.

Specialized Tools

Embed expert judgment into equipment or software practitioners already use.

Licensing

License structured cases, evaluation methods, or task-specific intelligence to manufacturers, training organizations, software companies, or robotics developers.

Robotics

Use accumulated expertise, demonstrations, sensing, and verified outcomes to automate bounded parts of physical work.

Do not assume the most technical business is the best business.

A profitable specialist service company with proprietary knowledge may be much more defensible than a generic AI application.

9. Test Whether the Knowledge Transfers

A collection is not valuable merely because it sounds impressive.

Test it against real performance.

Possible tests include:

Give a junior practitioner unfamiliar cases and compare performance with and without access to your material.
Ask experts to review the same case independently.
Have someone predict what the expert will do next before revealing the outcome.
Use historical cases while withholding the final diagnosis or result.
Track whether recommendations actually resolve the problem.
Measure time saved, mistakes avoided, quality improved, proficiency accelerated, or escalation reduced.

Then ask someone to pay.

People may compliment an interesting project for months. Asking for $500 changes the conversation.

The goal is evidence that the captured judgment changes behavior or outcomes, not merely that people find the archive interesting.

10. Build the Second Asset

The first asset is the accumulated expertise.

The second asset is the machinery that created it.

By the end of the experiment, ask whether you have become better at:

Prospecting

Finding domains where undercaptured judgment has real value.

Partnering

Earning trust and creating durable incentives for experts.

Capturing

Recognizing consequential moments and preserving enough context to understand them later.

Protecting

Maintaining confidentiality, rights, provenance, and control over valuable material.

Refining

Turning messy work into structured, validated cases.

Applying

Finding problems the knowledge can solve and customers willing to pay.

Testing

Determining whether the knowledge actually improves decisions or outcomes.

Repeating

Taking the method into a new domain and improving it through another cycle.

That combination is your knowledge-mining machinery.

The first domain teaches you what the machinery needs. The second tells you whether it is portable.

11. Run a Second Experiment Before You Declare Victory

Near the end of the 18 months, choose a small second domain.

Do not attempt to reproduce the entire first project. Run a compressed version.

Can you identify a promising expert faster?

Can you explain the proposition more clearly?

Can you negotiate rights earlier?

Can you recognize consequential decisions sooner?

Can you create useful cases with less wasted capture?

Can you protect and organize the material correctly from day one?

Can you identify a useful application faster?

If the answer is yes, you may have built something more important than a single archive.

You may have learned how to create knowledge assets repeatedly.

The second experiment is therefore a test of your profession, not just your process.

12. Know What Can Go Wrong

Professional Knowledge Mining has several recurring traps.

Mistaking Explanation for Expertise

People can give persuasive explanations that do not actually explain their decisions. Compare statements with behavior and outcomes.

Capturing Outdated Knowledge

Experts can be wrong. Equipment changes. Standards change. Materials change. Preserve dates, conditions, boundaries, and revision history.

Creating Surveillance

Workers may reasonably fear that cameras and AI are being used to monitor or replace them. Consent, incentives, permitted uses, and trust are part of the method.

Choosing the Wrong Hardware

Consumer smart glasses may be inappropriate or unsafe in some industrial environments. The device must fit the work.

Mining Knowledge That Is Already Cheap

If a public AI model, YouTube, manuals, and search already reproduce the expertise adequately, you may have chosen a weak target.

Look for judgment that is local, sensory, historical, rare, proprietary, or tightly connected to outcomes.

Building Before You Have Evidence

A polished application wrapped around shallow knowledge does not solve the problem.

The difficult part is still acquiring reliable expertise. Modern AI changes the economics of capture and reuse without eliminating the human challenge.

13. Where Robots Eventually Enter

Do not start by trying to automate an entire craft.

Start by helping humans perform it better.

If that succeeds, each job produces more evidence: what happened, what was tried, what worked, what failed, and where the boundaries are.

Then automate bounded tasks where perception, action, and success can be measured reliably.

A robot might begin with inspection. Then positioning. Then one controlled adjustment. Then another.

The advantage compounds when better judgment improves the work, the work creates new evidence, and the evidence improves the next decision.

Eventually, physical automation becomes one consumer of the knowledge asset rather than the reason the project existed in the first place.

The robot may be commercially available to everyone.

The accumulated judgment behind it may not be.

14. What Success Looks Like After 18 Months

Do not judge the experiment solely by whether you incorporated a startup.

A successful 18 months might leave you with:

a trusted relationship with one or more exceptional practitioners;
a body of well-controlled source material;
dozens or hundreds of structured cases;
a clear understanding of which judgment is genuinely scarce;
documented ownership and usage rights;
sensitivity rules for handling valuable material;
traceable provenance from cases back to evidence;
a working method for capturing and refining new cases;
evidence that the material helps someone perform better;
at least one small thing someone will pay for;
a second-domain test showing which methods transfer;
a much sharper idea of where the opportunity actually lies.

That is an extraordinary amount of leverage for a university student to create in two or three summers.

You may finish the experiment with a company.

You may finish with a profession.

Or you may finish with a way of seeing valuable human knowledge that most people still do not recognize as an asset.

Any of those would make the experiment worth running.